Soil moisture data obtained by inversion of Fengyun 3B remote sensing data, are widely used in drought monitoring and global climate change research, however, some regional data are missing in this data set, which reduces the application effect. Based on backpropagation neural network (BPNN), we established a filling method and filled the missing area with moderate resolution imaging spectroradiometer (MODIS) inversion products, including land surface temperature, normalized difference vegetation index, and albedo. We named it the multilayer BPNN filling algorithm. The algorithm consists of two neural network layers. The first network layer is used for the spatial scaling of MODIS inversion products, and the second network layer uses the scaling products to further generate soil moisture values. We compared the proposed method to a discrete cosine transform and partial least square (DCT-PLS) and a kriging using the same data set. The experiments demonstrate that our method could obtain good filling results in both homogeneous areas and areas with high data variations, whereas DCT-PLS and kriging could only get good filling results in homogeneous areas.
Convolutional neural network (CNN) models achieve state-of-the-art performance for natural image semantic segmentation. An approach for extracting vegetation from Gaofen-2 (GF-2) remote sensing imagery based on the CNN model is presented. We constructed a convolutional encoder neural networks (CENN) consisting of two layers. The first layer has two sets of convolutional kernels for extracting the features of farmland and woodland, respectively. The second layer consists of two encoders that use nonlinear functions to encode the learned features and map the encoding results to the corresponding category number. In the training stage, samples of farmland, woodland, and other lands are categorically used to train the CENN. After training is accomplished, the CENN would acquire enough ability to accurately extract farmland and woodland from GF-2 imagery. The CENN was trained on 36 GF-2 images and tested on three other GF-2 images. We compared the proposed method to a deep belief network, a fully convolutional network, and a DeepLab model using the same images. The experiments demonstrate that the proposed approach improves upon the accuracy of existing approaches. The average precision, recall, and kappa coefficient of the proposed approach were 0.91, 0.87, and 0.86, respectively. Thus, the proposed approach is proven to effectively extract vegetation from GF-2 imagery.
The "digital reservoir" is usually understood as describing the whole reservoir with digital information technology to make it serve the human existence and development furthest. Strictly speaking, the "digital reservoir" is referred to describing vast information of the reservoir in different dimension and space-time by RS, GPS, GIS, telemetry, remote-control and virtual reality technology based on computer, multi-media, large-scale memory and wide-band networks technology for the human existence, development and daily work, life and entertainment. The core of "digital reservoir" is to realize the intelligence and visibility of vast information of the reservoir through computers and networks. The dam is main building of reservoir, whose safety concerns reservoir and people's safety. Safety monitoring is important way guaranteeing the dam's safety, which controls the dam's running through collecting the dam's information concerned and developing trend. Safety monitoring of the dam is the process from collection and processing of initial safety information to forming safety concept in the brain. The paper mainly researches information collection and processing of the dam by digital means.